Financial analysis you can check
MazeCFO answers questions about your financials in plain language — and shows you where every number came from. Figures are produced by deterministic calculations against your own data, never written by a language model.
The problem with AI in finance
A language model asked to compute a gross margin will produce a plausible number. It may even be right. But you cannot audit it, you cannot reproduce it, and you cannot put it in a board pack — because there is no way to show where it came from.
MazeCFO is built the other way round. The model reads your question and decides which calculation to run. The calculation itself is deterministic code operating on your ledger, returning a value together with its inputs, its assumptions, and a trace of every step. The model explains the result; it does not produce it.
The practical difference: when a number looks wrong, you can find out why in about four clicks instead of re-deriving it in a spreadsheet.
What every answer carries
Data period and currency
Which months the figure covers and what currency it is denominated in, stated on the answer rather than assumed from context.
Sources
The specific import, spreadsheet range, or PDF page each input came from. Citations link to the exact cell or page, not the document.
Calculation trace
Every input and intermediate step, expandable inline. Reproducible by hand if you want to check it.
Assumptions
Stated explicitly — recognition basis, FX rate used, treatment of partial periods — instead of being buried in a model's reasoning.
Limitations
What was missing. If four days of bank feed are absent, the answer says so rather than quietly extrapolating.
Cost
Tokens and spend for the request, so AI usage is a line item you can see rather than a surprise at renewal.
Four products, one workspace
AI Financial Analyst
Management reporting, runway, budget versus actual, and the monthly close for teams without a dedicated analyst.
AI CFO
Forecasting, scenarios, 13-week cash flow, department budgets, and headcount planning.
AI Due Diligence
Data-room requests, findings tracking, and evidence linked back to the page it came from.
AI Hedge Fund Research
Securities, portfolios, positions, research notes, and investment theses with explicit assumptions and risks.
Built for data that cannot leak
Tenant isolation is enforced by PostgreSQL row-level security, not by application code remembering to filter. The application connects as a database role that cannot bypass those policies — it is neither a superuser nor the owner of any table, which are the two ways row-level security is silently defeated.
A query that forgets its tenant context raises an error rather than returning an empty result, because an empty result is indistinguishable from a genuinely empty table and would ship unnoticed.